lumber-mart

Skills for the lightspec spec pipeline: draft a feature spec into specs/NNN-slug/spec.md in a script-enforced format, refine its rough edges with a short interview, approve it on substance, and convert an accepted spec into a hew epic with one child issue per user story. research → ADR → spec → refine → approve → tracker.

4 skills

Review skills for code, tests, and documentation. Finds code design and correctness, coverage, and doc-structure issues that linters miss.

3 skills

Skills for researching, deciding, and enforcing: create-research records a question investigated with per-finding confidence and cited sources, create-adr writes Architectural Decision Records (ADRs) as standalone, append-only policy in MADR format that cite the research they rest on, and review-policy checks code changes against accepted ADRs.

3 skills

Skills for tracking work through the hew CLI, with GitHub or local trackers. File deduplicated review findings, implement issues test-first through a draft PR or committed branch, then review, verify, squash merge, and confirm closure with finish-issue.

3 skills

Skills for creating and maintaining operational runbooks with consistent structure and built-in maintenance feedback loops.

1 skill

Review skill files for agent-harness coupling that breaks or degrades them when a different coding agent (Claude Code, opencode, codex) runs them.

1 skill

Skill for burning down a hew epic or a set of issues across Orca-managed worktrees and supervised agents, driven by a tested script rather than by an agent following prose. The driver resolves the ready issues, creates one Orca worktree per issue, starts an implementer running work-issue, has a different agent review each delivered PR once and fix what it finds in that same session, files every finding as an issue for the paper trail, repairs merge conflicts and red CI, and merges PRs once CI is green. Runs unattended and stops only for a decision a human has to make.

1 skill